Other Very Rarely used qualifiers

Introduction

Qualifiers are described in 4 sections based on frequency of use.
  • general syntax
  • The
    • major qualifiers are !CONTINUE, !CONTRAST, !FCON, !MAXIT, !SUBSET, !SUM and XYplots ( !X !Y !G !JOIN )
  • Minor qualifiers
  • Special qualifiers
  • The qualifiers described here are other rarely required qualifiers.
    • They include
        Additional output: !CINV, !REPORT, !SCORE, !VRB
          Adjusting detection of singularities in the mixed model equations: !TOLERANCE,
         Setting covariate grouping levels: !FACPOINTS, !POLPOINTS,

    !CINV

    ASReml does not compute the whole C-inverse matrix but only sufficient to calculate the REML solution. The user may request that the part that is calculated be written to a .cii file. !CINV n prints the portion of the inverse of the coefficient matrix (C) pertaining to the nth term in the linear model, or if n is 1 or omitted, pertaining to all the random (sparse) terms. Because the model has not been defined when ASReml reads this line, it is up to the user to count the terms in the model to identify the portion of the inverse of the coefficient matrix to be printed. The option is ignored if the portion is not wholly in the SPARSE stored equations. The portion of the inverse is printed to a file with extension .cii. The sparse form of the matrix only is printed in the form
         i      j      Cij
    that is, elements of the C inverse that were not needed in the estimation process are not included in the file.

    Setting covariate grouping levels:

    The fac() and pol() model terms need to classify a covariate into discrete classes. This is usually done on the basis of unique covariate values. However, ASReml will actually consider close points as being the same. This is controlled by !FACPOINTS and !POLPOINTS qualifiers respectively.

    Imputation

    !IMPUTE [ n] is a special developmental option proposed for fitting large models by splitting the model, fitting the reduced models alternately after adjusting for effects in the other submodel, with added error. The submodels are defined using the !SM m qualifier in the model line. More

    !SM [ m] can be used as a data line qualifier in conjunction with !SM [ m] as a model qualifier to fit the selected submodel without added error.

    Transformed data scratch file.

    !NOSCRATCH forces ASReml to hold the data in memory. ASReml will usually hold the data on a scratch file rather than in memory. In large jobs, the system area where scratch files are held may not be large enough. A Unix system may put this file in the /tmp directory which may not have enough space to hold it. If ASReml crashes, it may leave large temporary files in /tmp which should be periodically checked and cleaned out.

    Standard output report

    !REPORT forces ASReml to attempt to produce the standard output report when there is a failure of the iteration algorithm. Usually no report is produced unless the algorithm has at least produced estimates for the fixed and random effects in the model. Note that residuals are not included in the output forced by this qualifier. This option is primarily intended to help debugging a job that is not converging properly.

    Exporting the AI matrix and Score

    !SCORE requests ASReml write the SCORE vector and the Average Information matrix to files basename.SCO and basename.AIM. The values written are from the last iteration.

    Singularity detection tolerance

    !TOLERANCE [s1/var> [ s2/var>]] modifies the ability of ASReml to detect singularities in the mixed model equations. This is intended for use on the rare occasions when ASReml detects singularities after the first iteration (they are not expected then), or it finds the variances of fixed effects are suspiciously large. ASReml will give a warning when the !TOLERANCE qualifier may be needed.

    Normally (when !TOLERANCE is not specified), a singularity is declared if the adjusted sum of squares of a covariable is less than eta or less than the uncorrected sum of squares times eta, where eta is 10-8 in the first iteration and 10-10 thereafter. The qualifier scales eta by 10si/sup> for the the first or subsequent iterations respectively, so that it is more likely an equation will be declared singular. Once a singularity is detected, the corresponding equation is dropped (forced to be zero) in subsequent iterations. If neither argument is supplied, 2 is assumed. If the second argument is omitted, it is given the value of the first.

    If the problem of later singularities arises because of the low coefficient of variation of a covariable, it would be better to centre and rescale the covariable. If the degrees of freedom are correct in the first iteration, the problem will be with the variance parameters and a different variance model (or variance constraints) is required.

    The .vrb file

    !VRB requests writing of the .vrb file. Previously, the default was to write the file but it is rarely needed. It contains the values of the residual variance, the fixed effects and the variance of the fixed effects in triangle rowwise order.

    Return to start